Autonomous Vehicle Safety

Ensure vehicle safety across the full AV stack with NVIDIA Halos.

Autonomous Vehicle Safety, From the Cloud to the Car

NVIDIA Halos is a full-stack, comprehensive safety system that unifies vehicle architecture, AI models, chips, software, tools, and services to ensure the safe development and deployment of autonomous vehicles (AVs) from cloud to car.

The system covers the full development lifecycle with design-time, deployment-time, and validation-time guardrails that collectively build safety and explainability into AI-based AV stacks. These guardrails are implemented using three powerful computers—NVIDIA DGX^™ for AI training, NVIDIA Omniverse^™ and Cosmos^™ for simulation, and NVIDIA AGX^™ for deployment. At the heart of the vehicle, NVIDIA Halos OS provides the unified software foundation necessary to bridge these AI capabilities with production-ready safety.

NVIDIA Halos complements existing industry-standard safety practices, while introducing unique elements for autonomous vehicles. This ensures regulatory compliance and advances safe and reliable AV stacks, together with NVIDIA’s Halos AI Systems Inspection Lab.

Halos is also extending its comprehensive safety framework beyond AVs to robotics, further enhancing the reliability and safety of intelligent systems.

NVIDIA Releases New AI Models and Developer Tools to Advance Autonomous Vehicle Ecosystem

To accelerate the development of next-generation AV architectures, NVIDIA released NVIDIA Cosmos Predict-2 — a new world foundation model with improved future world state prediction capabilities for high-quality synthetic data generation — as well as new developers tools.

NVIDIA Launches Halos, a Full-Stack, Comprehensive Safety System for Autonomous Vehicles

NVIDIA unifies vehicle architecture to AI models; chips, software, and tools to services for safely developing AVs from cloud to car.

18,600+

Engineering years invested in vehicle safety to date

21 Billion+

Safety transistors safety assessed

7,000,000

Lines of safety-assessed code

2,000,000

Daily end-to-end integration tests for validation

22,000+

Platform safety monitors

20,000+

Hours of safety test data

330+

Research papers published on AV safety

30+

Certificates and assessment reports issued

NVIDIA DGX

Design-time safety guardrails for built-in hardware/software safety and trustworthy development processes.

NVIDIA Omniverse With Cosmos

Validation-time guardrails for data generation, simulation, evaluation, and lifelong safety assurances.

NVIDIA DRIVE AGX™

Deployment-time guardrails for run-time monitoring and real-time introspection.

Platform Safety

The robust foundation for autonomous driving systems includes: 

• A System-on-a-Chip (SoC) that’s designed for safety, with hundreds of built-in safety mechanisms.
• NVIDIA Halos OS, a unified, three-layer safety foundation built on ASIL D certified NVIDIA DriveOS™. The architecture integrates Halos Core (safety operating system), Halos SDK (safety middleware), and Halos applications (safety applications) to deliver a production-ready environment for AV.
• A safety-assessed base platform that delivers the foundational safe computer needed to enable safe systems for all types of applications
• NVIDIA DRIVE Hyperion™, a diverse hardware platform that connects the SoC, OS, and sensors in a vehicle architecture. This enables a vehicle to safely execute contingency plans if needed.

Algorithmic Safety

Algorithmic AI safety spans:

• A diverse AV stack that combines a modular stack and NVIDIA Alpamayo reasoning VLA models for algorithmic AI safety.
• Training, simulation, and validation environments that use NVIDIA Omniverse and Cosmos platforms to build safe AVs.
• A separate safety dataset that ensures AV performance is tested against diverse, unbiased data.

Ecosystem Safety

Building a safer AV ecosystem includes:

• Continual improvements through a safety data flywheel, which continually learns from the road how to expand the set of operational design domains for safe deployment.
• Seamless integration of physically based and diverse sensor simulation into existing workflows to safely train, test, and validate AVs with the NVIDIA Omniverse Blueprint for AV Simulation.
• Open-source data, such as the NVIDIA Physical AI Dataset, to enable critical safety research throughout the industry.
• A growing list of partners using the NVIDIA Halos system, listed here.

ANSI National Accreditation Board

ANAB accredited the NVIDIA Halos AI Systems Inspection Lab as an ISO/IEC 17020 Inspection Body. NVIDIA is the first company accredited by ANAB for an inspection plan that combines cybersecurity, AI, and functional safety.

TÜV SÜD

TÜV SÜD certified the core NVIDIA hardware and software process to Automotive Safety Integrity Level (ASIL) D.  Under the ISO 26262 standard, NVIDIA DriveOS 6.0 is certified  ASIL D conformant and Thor-X SoC is assessed as ASIL D conformant. NVIDIA also received ISO/SAE 21434 Cybersecurity Process certification for its automotive system-on-a-chip, platform, and software engineering processes.

TÜV Rheinland

TÜV Rheinland performed an independent United Nations Economic Commission for Europe safety assessment of NVIDIA DRIVE AV related to safety requirements for complex electronic systems.

Alpamayo 1: Bridging Reasoning and Action Prediction for Generalizable Autonomous Driving in the Long Tail

Comprehensive evaluations with open-loop metrics, closed-loop simulation, and real-world vehicle tests demonstrate that Alpamayo 1 is state-of-the-art in multiple aspects (including reasoning, trajectory generation, alignment, safety, latency, and more).

Cosmos-Drive-Dreams: Scalable Synthetic Driving Data Generation With World Foundation Models

Collecting and annotating real-world data for safety-critical physical AI systems, such as Autonomous Vehicles (AVs), is time-consuming and costly. To address this challenge, we introduce the Cosmos-Drive-Dreams, a synthetic data generation (SDG) pipeline for generating challenging scenarios to facilitate downstream tasks such as perception and driving policy training.

Sim2Val: Using Correlation Across Test Platforms for Variance-Reduced Metric Estimation

Learning-based robotic systems demands rigorous validation to assure reliable performance. However, extensive real-world testing is often prohibitively expensive and, if conducted, may still yield insufficient data for high-confidence guarantees. That’s where Sim2Val comes in.

SafeVL: Driving Safety Evaluation Using Meticulous Reasoning in Vision Language Models

Safety remains a fundamental challenge in autonomous driving, with a key step being the development of a safety evaluator that can reliably identify unsafe (i.e., collision-prone) scenarios.

Enhancing Autonomous Driving Safety With Collision Scenario Integration

HydraSafe is a framework that addresses the challenge of ensuring autonomous vehicle safety in hazardous scenarios by improving data availability and planner robustness.

Safety Evaluation of Motion Plans Using Trajectory Predictors as Forward Reachable Set Estimators

The advent of end-to-end autonomy stacks—often lacking interpretable intermediate modules—has placed an increased burden on ensuring that the final output, i.e., the motion plan, is safe in order to validate the safety of the entire stack.

NVIDIA Alpamayo

Designed to handle real-world scenarios, Alpamayo accelerates safe Level 4 AV development by enabling reasoning-based autonomy through a family of open VLA models, physical AI datasets, and simulation frameworks.

Cars

Trucks

Robotaxis

Suppliers

Simulation

Sensors

Software

Mapping

1. News

2. Sessions

3. Live Streams